Photoshop: fix camera shake as well as focus blur

  • 2
  • Idea
  • Updated 5 years ago
  • Implemented
  • (Edited)
I can see 2 more features to add, which will be excellent combined together:

the equivalent of topaz infocus plugin (terrific plugin) to fix out-of-focus "blur"
smart kernel-based camera-shake "blur" fix such as adobe was going to offer in a future version of ps

*combine* these two, and you will have a powerful solution which can fix any kind of blur problem. blur problems are not just camera shake alone. there are 2 types of blur I have seen on the internet: out of focus blur and camera shake "blur".
there are a number of methods for handling camera shake, some even having university papers I think.

nobody likes a blurred photo. and professionals require images to be tack sharp.
Photo of Jim

Jim

  • 68 Posts
  • 1 Reply Like

Posted 6 years ago

  • 2
Photo of Lee Jay

Lee Jay

  • 990 Posts
  • 135 Reply Likes
Infocus is a bad model since it absolutely stinks. Focus Magic blows it away by a million to one. I think Adobe should concentrate on beating the performance of Focus Magic.
Photo of Jim

Jim

  • 68 Posts
  • 1 Reply Like
focus magic's out-of-focus blur handling technology is probably pretty good,but the motion blur handling as they call it (commonly known as camera shake) is not handled anywhere near as well as what was announced in an adobe conference. I guess it was staged, but if Adobe truly has fleshed out any kind of idea for converting an image to a kernel, then this project will be completable!

see http://forums.adobe.com/thread/105311... I posted here but found out copy and paste is disabled, so could not post here as well. :-(

http://cs.nyu.edu/~fergus/papers/debl...
http://www.mendeley.com/research/remo...
that is the paper for doing the blur kernel. it does not look perfect. but that may have depended on the sharpness or contrast (?) of the selection area maybe? more experimentation would be needed.
or another algorithm.

another kernel-based implementation is
http://www.adptools.com/en/deblurmyim...
http://www.windows7download.com/win7-...
but both kernel-based convolution results don't always look pretty, may introduce noise so image may need a denoiser plugin like topaz denoise.

Jim
Photo of Jim

Jim

  • 68 Posts
  • 1 Reply Like
here is the text from that other post.
one of the things I think about as an engineer is that in order for something like this to work, you should probably start with manuals adjustments to processing the the out-of-focus blur first. after you have finished with this, you can do the motion blur.

I would have to really think hard about it to see if there is an algorithm such as a kernel or something to convert an image into a path (there is,it' above). one problem is you are stuck with the image boundaries. if they weren't there, things would be so much easier (but it would also take infinitely long to process). the image boundaries MIGHT skew the kernel, depending on how you handle the boundaries (for example, usually you just assume those values are black or 0's, or you may choose a 50% gray, whatever fits best with the kernel you are working with).

adobe may be on to something there, but they only had half of the equation with the idea.

I used to tinker with image processing many yarns ago back for fun in BASICA/GWBASIC.

if you have the kernel
[-2 0 2]
[-2 0 2]
[-2 0 2]
which amplifies vertical lines, you would want to handle the boundaries by processing them as 0's.

NOTE TO ADOBE: it might start getting into Machine Vision technology when you start doing the path recognition. or AI.

if you can get the imge into a path kernel, it's essentially a bitmap of a path. you can trace that path using akima spline curves. (will lose 1-2 samples off each end), but the tracing program might just exist out there somewhere. just google "bitmap to akima spline curve". and akima spline curve has all the

points along its path, unlike a bezier. I like akima spline curves, I have been trying to get ps engineers to use them for a while for good reason. this is one of them.

there are also cubic splines as an alternative.

chances are that the curve will have rounded corners due to the weight of the camera.

now that you have the path,it's simply a matter of taking samples along the curve wherever you want. (usually in an array which you can linearly inerpolate across the points in the array for simplicity if you want)
Photo of Jim

Jim

  • 68 Posts
  • 1 Reply Like
hmm. after analyzing the results of the kernel-based solutions, I noticed they have ghosting (ringing?).
possibly a good solution would combine something like focus magic with adptools.com's solution?
maybe it's too early to tell yet. I would change the kernel algorithm from bezier curves to akima spline curves personally. might work better.
Photo of Jim

Jim

  • 68 Posts
  • 1 Reply Like
http://tv.adobe.com/watch/creative-cl...
the camera shake feature has come to CC. yay!